CN113109874B - Wave impedance inversion method using neural network and neural network system - Google Patents
Wave impedance inversion method using neural network and neural network system Download PDFInfo
- Publication number
- CN113109874B CN113109874B CN202110339626.5A CN202110339626A CN113109874B CN 113109874 B CN113109874 B CN 113109874B CN 202110339626 A CN202110339626 A CN 202110339626A CN 113109874 B CN113109874 B CN 113109874B
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- layer
- seismic
- seismic data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/40—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging
- G01V1/44—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging using generators and receivers in the same well
- G01V1/48—Processing data
- G01V1/50—Analysing data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/61—Analysis by combining or comparing a seismic data set with other data
- G01V2210/616—Data from specific type of measurement
- G01V2210/6169—Data from specific type of measurement using well-logging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Remote Sensing (AREA)
- Geophysics (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Acoustics & Sound (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The application relates to a wave impedance inversion method and a neural network system using a neural network, wherein the neural network takes N channels of seismic data consisting of ith channel of seismic data and a plurality of channels of seismic data adjacent to the ith channel of seismic data and ith channel of initial model data as input, and determines the ith channel of wave impedance data as output, and the method comprises the following steps: n parallel feature extraction layers, wherein each feature extraction layer is configured to extract time-series features of a piece of seismic data input thereto; the merging layer is configured to adaptively merge the time sequence characteristics output by the N characteristic extraction layers to obtain the time-space characteristics of the N seismic data; a regression layer configured to map spatiotemporal features from a feature domain to a target domain; an output layer configured to determine ith channel wave impedance data from the output of the regression layer and the ith channel initial model data. Therefore, the wave impedance inversion has higher precision, stronger continuity and better noise immunity, and can still keep good inversion effect when the initial model is not accurate.
Description
Technical Field
The application relates to the technical field of oil and gas exploration, in particular to a wave impedance inversion method using a neural network and a neural network system.
Background
The seismic wave impedance inversion is a method for recovering broadband wave impedance data from seismic data with limited frequency bandwidth by comprehensively utilizing the existing geological and logging data under the guidance of seismic data, and is widely applied to the aspects of reservoir qualitative and quantitative prediction in an oil and gas exploration stage, well pattern deployment in an oil and gas development stage, reserve calculation, dynamic monitoring of oil reservoirs and the like at present. Since the actual geophysical problem is complex and our understanding of it is often quite ambiguous, the models created when inverting the wave impedance are mostly approximate. The neural network can learn the knowledge hidden in the neural network from a large amount of existing data so as to establish a corresponding mathematical model, so that the neural network is very suitable for solving the problems that the knowledge background is not clear enough and the model is not accurate enough. A deep learning algorithm may therefore be applied in the wave impedance inversion.
There are many existing wave impedance inversion methods based on a deep neural network, such as a wave impedance inversion method based on a convolutional neural network and a wave impedance inversion method based on a closed-loop convolutional neural network. These networks learn the mapping between seismic data and inversion parameters directly from the training data set. When the underground structure is complex, the relationship between the seismic data and the inversion parameters is very complex, and the network generally cannot accurately express the relationship, and although the learning capacity of the network can be improved by increasing the depth of the network, the risk of overfitting of the network is increased due to the limitation of the data volume.
Disclosure of Invention
In order to solve the above technical problem or at least partially solve the above technical problem, the present application provides a wave impedance inversion method using a neural network and a neural network system.
In a first aspect, the present application provides a method of wave impedance inversion using a neural network, the method comprising: receiving N channels of seismic data, wherein the N channels of seismic data comprise the ith channel of seismic data and a plurality of channels of seismic data adjacent to the ith channel of seismic data; receiving the ith initial model data; extracting time sequence characteristics of N pieces of seismic data by N parallel characteristic extraction layers of a neural network; adaptively combining the time sequence characteristics output by the N characteristic extraction layers by a combining layer of the neural network to obtain the time-space characteristics of the N pieces of seismic data; mapping the spatiotemporal features from the feature domain to a target domain by a regression layer of the neural network; and determining the ith channel wave impedance data by an output layer of the neural network according to the output of the regression layer and the ith channel initial model data.
In some embodiments, before determining, by the output layer of the neural network, the ith channel wave impedance data from the output of the regression layer and the ith initial model data, the method further includes: and carrying out self-adaptive adjustment on the ith initial model data by a correction layer of the neural network.
In some embodiments, each feature extraction layer comprises: a global feature extraction layer configured to extract global features in the input seismic traces; a local feature extraction layer configured to extract local features in the input seismic traces; and/or, the merging layer is a linear connection layer with zero offset; and/or, the regression layer comprises: a set of serial deconvolution blocks configured to upsample the output of the combining layer; a gated round-robin unit and a fully connected layer configured to map the upsampled data from the feature domain to the target domain.
In certain embodiments, the correction layer is a linear connection layer and a Tanh activation function.
In a second aspect, the present application provides a neural network system for wave impedance inversion, the neural network system being implemented by one or more computers, the neural network system being configured to take as input N-channel seismic data composed of an ith channel seismic data and multiple channels of seismic data adjacent to the ith channel seismic data, and an ith channel initial model data, and determine as output the ith channel wave impedance data, the neural network system comprising: n parallel feature extraction layers, wherein each feature extraction layer is configured to extract time-series features of a piece of seismic data input thereto; the merging layer is configured to adaptively merge the time sequence characteristics output by the N characteristic extraction layers to obtain the time-space characteristics of the N pieces of seismic data; a regression layer configured to map spatiotemporal features from a feature domain to a target domain; an output layer configured to determine ith channel wave impedance data from the output of the regression layer and the ith channel initial model data.
In some embodiments, the neural network system further includes: and the correction layer is positioned before the output layer and is configured to perform adaptive adjustment on the ith track of initial model data.
In some embodiments, each feature extraction layer comprises: a global feature extraction layer configured to extract global features in input seismic traces; a local feature extraction layer configured to extract local features in the input seismic traces; and/or, the merging layer is a linear connection layer with zero offset; and/or, the regression layer comprises: a set of serial deconvolution blocks configured to upsample the output of the combining layer; a gated round-robin unit and a fully connected layer configured to map the upsampled data from the feature domain to the target domain.
In certain embodiments, the correction layer is a linear connection layer and a Tanh activation function.
In a third aspect, the present application provides a method of training a neural network for wave impedance inversion, comprising: receiving first input data and second input data, wherein the first input data comprises: n seismic data and ith initial model data, wherein the N seismic data are composed of ith seismic data and multiple adjacent seismic data; the second input data includes: seismic data at the well site and wave impedance data at the well site; determining ith channel wave impedance data by the inversion neural network according to the first input data, determining synthesized first seismic data by the forward neural network according to the ith channel wave impedance data output by the inversion neural network, and determining synthesized second seismic data by the forward neural network according to the ith channel wave impedance data input at the well position; determining a first mean square error between the input ith seismic data and the synthesized first seismic data, determining a second mean square error between wave impedance data output by the inversion neural network and the input wave impedance data of the ith seismic data at the well position, and determining a third mean square error between the synthesized second seismic data of the ith seismic data and the input seismic data at the well position; the parameters of the inverse neural network and the forward neural network are updated using a first mean square error, the parameters of the inverse neural network are updated using a second mean square error, and the parameters of the forward neural network are updated using a third mean square error.
In some embodiments, the above inverse neural network comprises: n parallel feature extraction layers, wherein each feature extraction layer is configured to extract time-series features of a piece of seismic data input thereto; the merging layer is configured to adaptively merge the time sequence characteristics output by the N characteristic extraction layers to obtain the time-space characteristics of the N pieces of seismic data; a regression layer configured to map spatiotemporal features from a feature domain to a target domain; an output layer configured to determine ith channel wave impedance data from an output of the regression layer and ith channel initial model data; and/or, the forward neural network comprises: a set of one-dimensional convolutional layers in series and an activation function.
Compared with the prior art, the technical scheme provided by the embodiment of the application has the following advantages: the method provided by the embodiment of the application has the advantages of higher wave impedance inversion precision, stronger continuity and better noise immunity, and can still keep better inversion effect under the condition that the initial model is inaccurate.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive exercise.
Fig. 1 is a block diagram of a structure of an embodiment of a wave impedance inversion system according to an embodiment of the present disclosure;
FIG. 2 is a flow chart of an embodiment of a method for wave impedance inversion using a neural network according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of an embodiment of a system for training a neural network according to an embodiment of the present disclosure;
FIG. 4 is a flowchart of one embodiment of a method for training a neural network for wave impedance inversion according to an embodiment of the present disclosure;
FIG. 5 is a comparison graph of a wave impedance inversion result and a true wave impedance on a Marmousi2 model by applying the inversion method provided by the embodiment of the application;
FIG. 6 is a graph showing the inversion result of the wave impedance of an actual data; and
fig. 7 is a hardware schematic diagram of an implementation manner of a computer device according to an embodiment of the present application.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In the following description, suffixes such as "module", "part", or "unit" used to indicate elements are used only for facilitating the description of the present invention, and have no particular meaning in themselves. Thus, "module", "component" or "unit" may be used mixedly.
Wave impedance inversion using neural networks
The embodiment of the application provides a neural network for wave impedance inversion. Because the logging information and the geological structure information are generally comprehensively utilized when the initial model is established, the initial model can reflect the underground real structure to a certain extent. And the underground structure has certain correlation, the correlation is related to the distance, the closer the distance is, the stronger the correlation is, and conversely, the weaker the correlation is. Since the depth of a seismic section is often expressed as a temporal depth, the spatial correlation of the subsurface formations over the seismic section is manifested as a temporal correlation of the seismic traces in the longitudinal direction and a spatial correlation between the central trace and the adjacent traces in the transverse direction. Therefore, the relationship between the subsurface true wave impedance and the initial model and seismic data is expressed by equation (1) in the embodiment of the present application:
in the formula (1), mi,tThe real wave impedance value at the t moment on the ith channel is obtained;
the wave impedance value at the t moment on the ith channel in the initial model is obtained; { xi-k,t,…,xi-1,t,xi,t,…,xi+k,tThe amplitude sequence of the ith channel and the adjacent 2k channels of seismic data at the time t is shown, and it should be understood that in some embodiments, the amplitude sequence is not limited to the 2k channels of seismic data adjacent to the ith channel, and any multiple channels are possible; g (-) is an initial model modification function used to modify the initial model to reduce the difference between the initial model and the wave impedance of the subsurface real, it being understood that the initial model may not be modified in some embodiments, and the embodiments of the present application are not limited thereto; f (-) is a residual function that is used to represent the error value between the true wave impedance value and the wave impedance value in the initial model.
Based on the relationship between the true wave impedance and the initial model and the seismic data in the formula (1), the neural network model is constructed to learn the mapping relationship.
Fig. 1 is a block diagram of a structure of an implementation manner of a wave impedance inversion system provided in an embodiment of the present application, and as shown in fig. 1, the wave impedance inversion system includes: a neural network input 110, an inverse neural network system 120, and a neural network output 130.
The neural network input 110 includes: n seismic data, wherein the N seismic data comprise the ith seismic data and a plurality of seismic data adjacent to the ith seismic data; and the ith initial model data. As shown in fig. 1, the N-channel seismic data is the ith channel seismic data and 2k channels of seismic data adjacent to the ith channel seismic data, wherein k channels of seismic data are respectively located on two sides. It should be understood that the embodiment of the present application is not limited thereto, and multiple seismic data adjacent to the ith seismic data are all feasible. The neural network output 130 is the ith channel impedance data, which is the predicted value of the ith channel impedance.
The inverse neural network system 120, as shown in fig. 1, the inverse neural network system 120 includes: n feature extraction layers 121, a merging layer 122, a regression layer 123, a correction layer 124, and an output layer 125 are juxtaposed.
Each feature extraction layer 121 is configured to extract a time-series feature of a piece of seismic data input thereto. Each feature extraction layer 121 processes one path of seismic data, N parallel feature extraction layers 121 process N paths of seismic data in parallel, as shown in fig. 1, each feature extraction layer is 2k +1 path of seismic data, and the neural network system 120 includes 2k +1 feature extraction layers. The timing feature represents the correlation of the seismic traces in time.
In some embodiments, as shown in FIG. 1, each feature extraction layer 121 includes a global feature extraction layer 1211 and a local feature extraction layer 1212, where the global feature extraction layer 1211 is configured to extract global features of the input seismic traces and the local feature extraction layer 1212 is configured to extract local features in the input seismic traces. As an example, the global feature extraction layer 1211 is composed of a set of serial bidirectional gated round-robin units (GRUs); the local feature extraction layer 1212 includes: a set of parallel rolling blocks with different expansion coefficients, configured to extract local features of different scales in seismic traces; a fully connected layer and a convolution block configured to combine the extracted local features. It should be understood that other structures capable of extracting the correlation of the seismic traces in time are also conceivable in the embodiment of the present application, and the embodiment of the present application is not limited thereto.
Referring to fig. 1, each feature extraction layer 121 outputs time series features corresponding to its input seismic traces, and N feature extraction layers 121 output N time series features corresponding to N seismic traces. And the merging layer 122 is configured to adaptively merge the time sequence features output by the N feature extraction layers 121 to obtain the time-space features of the N seismic data. The spatiotemporal features represent not only the temporal correlation of a seismic trace, but also the spatial correlation between a seismic trace and an adjacent seismic trace. In some embodiments, the merging layer 122 is a linear connection layer with zero offset, but this is not limited by the embodiments of the present application.
The neural network input 110 provides time sequence characteristics through the N characteristic extraction layers 121, and outputs space-time characteristics through the merging layer 122, and the space-time characteristics are used as the input of the regression layer 123. A regression layer 123 configured to map spatio-temporal features from the feature domain to the target domain. In certain embodiments, the regression layer 123 includes: a set of serial deconvolution blocks configured to upsample the output of the merging layer 122 to a preset sample rate; a GRU and a full connectivity layer configured to map the upsampled data from the feature domain to the target domain.
The N parallel feature extraction layers 121, the merging layer 122 and the regression layer 123 complete the space-time sequence residual error modeling of the network, and the residual error function f (-) in the formula (1) is realized. The output of the regression layer 123 and the i-th pass initial model data are input together to the output layer 125. An output layer 125 configured to determine ith channel wave impedance data from the output of the regression layer 123 and the ith channel initial model data. In some embodiments, the output layer 125 adds the i-channel initial model data to the output of the regression layer 123 to obtain the i-th channel impedance data, but the embodiments of the present application are not limited thereto.
In some embodiments, as shown in FIG. 1, the i-th initial model data is further modified by a modification layer 124 located before the output layer 125 before being input into the output layer. The correction layer 124 is configured to adaptively adjust the ith pass of initial model data, and implement the correction function g (-) in equation (1). In certain embodiments, the correction layer 124 includes a linear connection layer and a Tanh activation function.
As a preferred example, the global feature extraction layer 1211 is comprised of a set of serial bidirectional GRUs to extract global features in the input seismic traces. The local feature extraction layer 1212 first extracts local features of different scales in the input seismic traces using a set of parallel volume blocks with different expansion coefficients, and then combines the local features using the fully connected layer and the volume blocks. The global feature extraction layer 1211 and the local feature extraction layer 1212 constitute a time series feature extraction module of the network, which is capable of extracting time series features of the input seismic traces. And extracting the time sequence characteristics of the ith seismic data and the adjacent 2k seismic data by applying 2k +1 time sequence characteristic extraction modules, and then performing self-adaptive combination on the extracted characteristics by using a combination layer 122 to obtain the time-space characteristics of the seismic data, wherein the combination layer 122 is composed of a linear connecting layer with the bias of zero. The regression layer 123 first upsamples the output of the merging layer to a predetermined sampling rate (e.g., the same sampling rate as the label data during training) using a set of concatenated deconvolution blocks, and then maps the upsampled data from the feature domain to the target domain using a GRU and a full-link layer. The regression layer 123 and the spatio-temporal feature extraction module constitute a spatio-temporal sequence residual error modeling module of the network. The correction layer 124 performs adaptive adjustment on the ith initial model data by using a linear connection layer and a Tanh activation function. Finally, the output of the space-time sequence residual modeling module is added to the output of the correction layer 124 to obtain the predicted wave impedance data.
The embodiment of the present application provides a wave impedance inversion method using a neural network, and as shown in fig. 2, the wave impedance inversion method includes steps S202 to S212.
Step S202, receiving N channels of seismic data, wherein the N channels of seismic data comprise the ith channel of seismic data and a plurality of channels of seismic data adjacent to the ith channel of seismic data;
step S204, the ith channel of initial model data is received.
And S206, extracting the time sequence characteristics of the N seismic data by the N parallel characteristic extraction layers of the neural network.
And S208, combining the time sequence characteristics output by the N characteristic extraction layers in a self-adaptive manner by the combination layer of the neural network to obtain the space-time characteristics of the N seismic data.
Step S210, the space-time characteristics are mapped to a target domain from a characteristic domain by a regression layer of the neural network.
And step S212, determining the ith channel wave impedance data by the output layer of the neural network according to the output of the regression layer and the ith channel initial model data.
It should be understood that although the sequence numbers of the steps are labeled in fig. 2, this is not a limitation on the order in which the steps are performed. For example, step S204 may be performed at any step prior to step S212.
In some embodiments, the step S212 further includes: and carrying out self-adaptive adjustment on the ith initial model data by a correction layer of the neural network. In certain embodiments, the correction layer is a linear connection layer and a Tanh activation function.
In some embodiments, the step S206 includes: and extracting global features in the input seismic channels by a global feature extraction layer of the feature extraction layer, and extracting local features in the input seismic channels by a local feature extraction layer of the feature extraction layer.
In some embodiments, the merged layer used in step S208 above is a linearly connected layer with a zero offset.
In some embodiments, in step S210, the output of the merging layer is first up-sampled by a set of serial deconvolution blocks of the regression layer, and then the up-sampled data is mapped from the feature domain to the target domain by the gated cyclic unit and the full-link layer of the regression layer.
In some embodiments, in step S212, the output of the regression layer and the ith channel initial model data are added by the output layer of the neural network to obtain the ith channel impedance data.
In the embodiment of the present application, the neural network may refer to the network structure shown in fig. 1, which is not described herein again.
As a preferred example, global features in the input seismic traces are extracted from a set of serial bidirectional GRUs, local features of different scales in the input seismic traces are extracted using a set of parallel rolling blocks with different coefficients of expansion, and then the local features are combined using fully-connected layers and rolling blocks, thereby extracting time-series features of the input seismic traces. And extracting time sequence characteristics of the ith channel of seismic data and the adjacent 2k channels of seismic data, and then carrying out self-adaptive combination on the extracted characteristics by using a combination layer so as to obtain the space-time characteristics of the seismic data, wherein the combination layer is composed of a linear connection layer with zero offset. The output of the merging layer is upsampled to a predetermined sampling rate (e.g., the same sampling rate as the label data during training) using a set of serial deconvolution blocks, and then the upsampled data is mapped from the feature domain to the target domain using a GRU and a full-link layer. And (4) performing self-adaptive adjustment on the ith initial model data by adopting a linear connection layer and a Tanh activation function. And finally, adding the output of the regression layer and the output of the correction layer to obtain predicted wave impedance data.
Training of neural networks
In consideration of the problem of label data shortage in actual exploration, the neural network is trained in a semi-supervised learning mode in the embodiment of the application. Fig. 3 is a schematic diagram of an embodiment of a system for training a neural network according to an embodiment of the present application, and as shown in fig. 3, the system includes: a first input 310, a second input 320, an inverse neural network 330, and a forward neural network 340.
As shown in fig. 3, the first input 310 is used to input first input data, where the first input data includes N-th channel seismic data composed of the ith channel seismic data and multiple channels of seismic data adjacent to the ith channel seismic data, and the ith channel initial model data, and is shown in fig. 3 as the ith channel seismic data and k channels of seismic data on both sides centered on the ith channel seismic data, which are 2k +1 channels of seismic data in total.
As shown in FIG. 3, the second input 320 is for inputting second input data, wherein the second input data includes wave impedance data 321 at the well site and seismic data 322 at the well site. The wave impedance data 321 at the well site is the actual wave impedance value (also referred to as input wave impedance data), and the seismic data 322 at the well site is the actual seismic data (also referred to as input seismic data).
As shown in fig. 3, the inverse neural network 330 is configured to determine ith trace impedance data 331 corresponding to the ith trace of seismic data from the first input data. The forward neural network 340 is configured to determine a synthesized ith trace of seismic data 341a from the output of the inverse neural network 330 (ith trace impedance data 331) and to determine a synthesized ith trace of seismic data 341b from the input ith trace impedance data 321.
As shown in fig. 3, the system further includes: a first error determination module 350 configured to determine a first mean square error (l) between the input ith trace of seismic data and the synthesized ith trace of seismic data 341aseismic) (ii) a A second error determination module 360 configured to determine a second mean square error (l) between the wave impedance data 331 and 321 at the wellsite for the ith tracewell) (ii) a A third error determination module 370 configured to determine a third mean square error (l'well)。
As shown in fig. 3, the inverse neural network 330 internal parameters are updated using the first and second mean square errors, and the forward neural network 340 internal parameters are updated using the first and third mean square errors.
In the embodiment of the present application, the inverse neural network 330 may refer to the network structure shown in fig. 1, which is not described herein again. The forward neural network 340 simulates the convolution model forward process from a set of one-dimensional convolution layers in series and activation functions.
The embodiment of the application provides a method for training a neural network for wave impedance inversion, which comprises steps S402 to S416 as shown in FIG. 4.
Step S402, receiving first input data and second input data.
The first input data comprises N channels of seismic data and ith channel of initial model data, wherein the N channels of seismic data are formed by ith channel of seismic data and multiple channels of seismic data adjacent to the ith channel of seismic data; the second input data includes: seismic data at the well site and wave impedance data at the well site.
Step S404, the ith channel wave impedance data is determined by the inverse neural network according to the first input data.
Step S406, the forward neural network determines the synthesized first seismic data according to the ith channel wave impedance data output by the inverse neural network.
And step S408, determining the synthesized second seismic data by the forward neural network according to the input ith channel wave impedance data at the well position.
Step S410, a first mean square error between the input ith trace of seismic data and the synthesized first seismic data is determined.
Step S412, determining a second mean square error between the wave impedance data output by the inversion neural network and the wave impedance data input at the ith channel at the well position.
Step S414, determining a third mean square error between the second seismic data synthesized by the ith track and the input seismic data at the well position.
Step S416, parameters of the inverse neural network and the forward neural network are updated.
Wherein the parameters of the inverse neural network and the forward neural network are updated using a first mean square error, the parameters of the inverse neural network are updated using a second mean square error, and the parameters of the forward neural network are updated using a third mean square error.
In some embodiments, the above-described inverse neural network comprises: n juxtaposed feature extraction layers, wherein each feature extraction layer is configured to extract a time-series feature of a trace of seismic data input thereto; the merging layer is configured to adaptively merge the time sequence characteristics output by the N characteristic extraction layers to obtain the time-space characteristics of the N pieces of seismic data; a regression layer configured to map spatiotemporal features from a feature domain to a target domain; an output layer configured to determine ith channel wave impedance data from the output of the regression layer and the ith channel initial model data.
In some embodiments, the inverse neural network further comprises a correction layer configured to adaptively adjust the ith pass initial model data. And an output layer configured to determine ith channel wave impedance data from the output of the regression layer and the corrected ith channel initial model data (output of the correction layer).
In the embodiment of the present application, the inverse neural network may refer to the network structure shown in fig. 1, which is not described herein again.
In some embodiments, the forward neural network comprises a series of one-dimensional convolution layers and activation functions to simulate a convolution model forward process.
As a preferred example, the global feature extraction layer of the inverse neural network consists of a set of serial bidirectional GRUs (gated-round units) to extract global features in the input seismic traces. The local feature extraction layer of the inversion neural network firstly extracts local features of different scales in input seismic channels by using a group of parallel volume blocks with different expansion coefficients, and then combines the local features by using a full connection layer and the volume blocks. The time sequence feature extraction module of the inverse neural network is formed by the two modules, and the time sequence feature extraction module can extract the time sequence feature of the input seismic channel. And (3) extracting the time sequence characteristics of the ith seismic data and the adjacent 2k seismic data by applying 2k +1 time sequence characteristic extraction modules, and then carrying out self-adaptive combination on the extracted characteristics by using a combination layer of an inverse neural network so as to obtain the space-time characteristics of the seismic data, wherein the combination layer consists of a linear connecting layer with zero offset. The regression layer of the inverse neural network first up-samples the output of the merging layer using a set of serial deconvolution blocks to have the same sampling rate as the tag data, and then maps the up-sampled data from the feature domain to the target domain using a GRU and a full-link layer. The time-space sequence residual error modeling module of the inverse neural network is formed by the time-space characteristic extraction module and the time-space sequence residual error modeling module. And a correction layer of the inversion neural network adopts a linear connection layer and a Tanh activation function to perform self-adaptive adjustment on the ith initial model data. And finally, adding the output of the space-time sequence residual error modeling module and the output of the correction layer to obtain the wave impedance data predicted by the network. The forward neural network simulates the forward process of a convolution model from a set of one-dimensional convolution layers in series and activation functions. During training, updating internal parameters of the inverse neural network and the forward neural network is comprehensively influenced by the following two processes:
1) The inversion neural network predicts the wave impedance data of the ith channel according to the input ith channel of seismic data, the adjacent 2k channels of seismic data and the ith channel of initial model data, and then inputs the predicted wave impedance data of the ith channel into the forward neural network to obtain the synthesized seismic data. By calculating the mean square error l of the synthesized ith trace seismic data and the input ith trace seismic dataseismicTo update the internal parameters of the inverse neural network and the forward neural network. All seismic traces participate in this process.
2) The inversion neural network predicts the wave impedance of the well position according to the input seismic data of the well position, 2k channels of seismic data adjacent to the well position and initial model data of the well position, and calculates the mean square error l of the predicted wave impedance data of the well position and the actual logging wave impedance datawellTo update parameters in the inverse neural network. The forward neural network predicts the corresponding seismic record according to the wave impedance in the actual logging, and calculates the mean square error l 'of the seismic data at the well position and the seismic data predicted by the forward neural network'wellTo update parameters in the forward neural network. Only seismic traces at the well site will participate in this process.
In the preferred example, the loss functions shown in equations (2) and (3) are constructed and Adam optimizer is adopted to update learnable parameters in the inverse neural network and the forward neural network:
wherein xi,tIs the input ith trace seismic data，
Is the synthetic ith trace seismic data output by the forward model,
is the wave impedance data in the well,
is the wave impedance data at the well site predicted by the inverse neural network,
is the seismic data at the well site,
is the seismic data at the well site predicted by the forward neural network, L (-) is a mean square error function defined as shown in formula (4), and α and β are weight coefficients whose values can be adjusted according to the quality of the seismic data and the well logging data.
Fig. 5 is a comparison between a wave impedance inversion result and a true wave impedance on a Marmousi2 model by applying the inversion method proposed in the embodiment of the present application, where (a) the true wave impedance, (b) the wave impedance inversion result, and (c) an error between the true wave impedance and a predicted wave impedance. FIG. 6 shows the inversion of the wave impedance of actual data. It can be seen that the inversion result based on the method provided by the embodiment of the application has a smaller error with the true wave impedance on the model data, the inversion result has a higher resolution in the actual data, and the inversion result is substantially consistent with the variation trend of the wave impedance of the test well at the test well.
According to the method and the device, a space-time sequence residual error modeling network is constructed, the network takes an initial model as an initial value, the initial model is continuously corrected in the learning process, the residual error between the corrected initial model and inversion parameters is learned, and the space-time characteristics of data are fully considered in the residual error learning process. In consideration of the problem of label data shortage in actual exploration, the network model is trained in a semi-supervised learning mode. The trained network may predict wave impedance data from the initial model data and the seismic data.
Compared with the prior art, semi-supervised learning seismic inversion in the related technology can well mine information in logging data and seismic data, the utilization rate of an initial model is poor, and the inversion of a network is generally restricted by only using the initial model as a label, so that the network cannot well utilize structural information and rich low-frequency information contained in the initial model.
In addition, the subsurface formations have a certain correlation in space, which is reflected in the seismic section as a spatial correlation in the lateral direction and a temporal correlation in the longitudinal direction. The deep learning inversion neural network in the related technology is sufficient for mining the time correlation in the longitudinal direction of the seismic data, but does not consider the spatial correlation of the seismic data in the transverse direction, so that the transverse continuity of the network prediction result is poor.
The embodiment of the application also provides computer equipment. The computer device 20 of the present embodiment includes at least but is not limited to: a memory 21, a processor 22, which may be communicatively coupled to each other via a system bus, as shown in FIG. 7. It is noted that fig. 7 only shows a computer device 20 with components 21-22, but it is to be understood that not all shown components are required to be implemented, and that more or fewer components may be implemented instead.
In the present embodiment, the memory 21 (i.e., a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, and the like. In some embodiments, the storage 21 may be an internal storage unit of the computer device 20, such as a hard disk or a memory of the computer device 20. In other embodiments, the memory 21 may also be an external storage device of the computer device 20, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like, provided on the computer device 20. Of course, the memory 21 may also include both internal and external storage devices of the computer device 20. In this embodiment, the memory 21 is generally used for storing an operating system installed in the computer device 20 and various application software, such as program codes of a method of wave impedance inversion. Further, the memory 21 may also be used to temporarily store various types of data that have been output or are to be output.
Processor 22 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 22 is typically used to control the overall operation of the computer device 20. In this embodiment, the processor 22 is configured to execute program code stored in the memory 21 or program code for processing data, such as a method of wave impedance inversion, to implement the method of wave impedance inversion.
The present embodiments also provide a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an App application mall, etc., on which a computer program is stored, which when executed by a processor, implements corresponding functions. The computer-readable storage medium of the present embodiment is for storing a program for wave impedance inversion, which when executed by a processor, implements the steps of the method for wave impedance inversion.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention or portions thereof contributing to the prior art may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods according to the embodiments of the present invention.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (7)
1. A wave impedance inversion method using a neural network, the wave impedance inversion method comprising:
receiving N channels of seismic data, wherein the N channels of seismic data comprise the ith channel of seismic data and a plurality of channels of seismic data adjacent to the ith channel of seismic data;
receiving the ith initial model data;
extracting time sequence characteristics of the N pieces of seismic data by N parallel characteristic extraction layers of the neural network;
adaptively combining the time sequence characteristics output by the N characteristic extraction layers by a combining layer of the neural network to obtain the space-time characteristics of the N seismic data, wherein the space-time characteristics not only represent the correlation of seismic channels in time, but also represent the spatial correlation between the seismic channels and adjacent seismic channels;
mapping, by a regression layer of the neural network, the spatiotemporal features from a feature domain to a target domain;
determining, by an output layer of the neural network, ith channel wave impedance data according to an output of the regression layer and the ith channel initial model data;
before determining the ith channel wave impedance data according to the output of the regression layer and the ith channel initial model data by the output layer of the neural network, the method further comprises the following steps: and carrying out self-adaptive adjustment on the ith initial model data by a correction layer of the neural network, continuously correcting the initial model in a learning process by taking the initial model as an initial value, learning a residual error between the corrected initial model and an inversion parameter, and fully considering the space-time characteristics of the data in the residual error learning process.
2. The wave impedance inversion method according to claim 1,
each feature extraction layer comprising: a global feature extraction layer configured to extract global features in the input seismic traces; a local feature extraction layer configured to extract local features in the input seismic traces; and/or
The merging layer is a linear connection layer with zero offset; and/or
The regression layer includes: a set of serial deconvolution blocks configured to upsample the output of the combining layer; a gated round-robin unit and a fully connected layer configured to map the upsampled data from the feature domain to the target domain.
3. The wave impedance inversion method of claim 1, wherein the correction layer is a linear connection layer and a Tanh activation function.
4. A neural network system for wave impedance inversion, the neural network system being implemented by one or more computers, the neural network system being configured to take as input N-th seismic data composed of an ith seismic data and a plurality of seismic data adjacent thereto, and ith initial model data, and determine as output ith wave impedance data, wherein the neural network system includes:
n parallel feature extraction layers, wherein each feature extraction layer is configured to extract time-series features of a piece of seismic data input thereto;
a merging layer configured to adaptively merge the time sequence features output by the N feature extraction layers to obtain a spatio-temporal feature of the N seismic data, wherein the spatio-temporal feature represents not only the correlation of a seismic trace in time but also the spatial correlation between the seismic trace and an adjacent seismic trace;
a regression layer configured to map the spatiotemporal features from a feature domain to a target domain;
an output layer configured to determine ith channel impedance data from an output of the regression layer and the ith initial model data;
the neural network system further includes: and the correction layer is positioned in front of the output layer and is configured to perform adaptive adjustment on the ith initial model data, the initial model is used as an initial value, the initial model is continuously corrected in the learning process, the residual between the corrected initial model and the inversion parameters is learned, and the spatio-temporal characteristics of the data are fully considered in the residual learning process.
5. The neural network system of claim 4,
each feature extraction layer comprising: a global feature extraction layer configured to extract global features in input seismic traces; a local feature extraction layer configured to extract local features in input seismic traces; and/or
The merging layer is a linear connection layer with zero offset; and/or
The regression layer includes: a set of serial deconvolution blocks configured to upsample the output of the combining layer; a gating cycle unit and a fully-connected layer configured to map the upsampled data from the feature domain to the target domain.
6. The neural network system of claim 4, wherein the modified layers are linear connection layers and Tanh activation functions.
7. A method of training a neural network for wave impedance inversion, comprising:
receiving first input data and second input data, wherein the first input data comprises: n seismic data and ith initial model data, wherein the N seismic data are composed of ith seismic data and multiple adjacent seismic data; the second input data comprises: seismic data at the well site and wave impedance data at the well site;
determining ith channel wave impedance data by the inversion neural network according to the first input data, determining synthesized first seismic data by the forward neural network according to the ith channel wave impedance data output by the inversion neural network, and determining synthesized second seismic data by the forward neural network according to the ith channel wave impedance data input at the well position;
determining a first mean square error between the input ith trace of seismic data and the synthesized first seismic data, determining a second mean square error between the wave impedance data output by the inverse neural network and the input wave impedance data for the ith trace at the well site, and determining a third mean square error between the synthesized second seismic data for the ith trace at the well site and the seismic data at the well site;
updating parameters of the inverse neural network and the forward neural network using the first mean square error, updating parameters of the inverse neural network using the second mean square error, and updating parameters of the forward neural network using the third mean square error;
the inverse neural network comprises: n parallel feature extraction layers, wherein each feature extraction layer is configured to extract time-series features of a piece of seismic data input thereto; a merging layer configured to adaptively merge the time sequence features output by the N feature extraction layers to obtain a spatio-temporal feature of the N seismic data, wherein the spatio-temporal feature represents not only the correlation of a seismic trace in time but also the spatial correlation between the seismic trace and an adjacent seismic trace; a regression layer configured to map the spatiotemporal features from a feature domain to a target domain; an output layer configured to determine ith channel impedance data from an output of the regression layer and the ith initial model data; a correction layer located before the output layer, the correction layer configured to perform adaptive adjustment on the ith initial model data, with the initial model as an initial value, continuously correcting the initial model in a learning process and learning a residual between the corrected initial model and an inversion parameter, and fully considering the spatiotemporal characteristics of the data in the residual learning process; and/or
The forward neural network comprises: a set of one-dimensional convolutional layers in series and an activation function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110339626.5A CN113109874B (en) | 2021-03-30 | 2021-03-30 | Wave impedance inversion method using neural network and neural network system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110339626.5A CN113109874B (en) | 2021-03-30 | 2021-03-30 | Wave impedance inversion method using neural network and neural network system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113109874A CN113109874A (en) | 2021-07-13 |
| CN113109874B true CN113109874B (en) | 2022-11-01 |
Family
ID=76712697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110339626.5A Active CN113109874B (en) | 2021-03-30 | 2021-03-30 | Wave impedance inversion method using neural network and neural network system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113109874B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113608264B (en) * | 2021-07-30 | 2023-06-09 | 成都理工大学 | Joint pre-stack elastic inversion parameter and depth network target inversion method |
| CN114063169B (en) * | 2021-11-10 | 2023-03-14 | 中国石油大学(北京) | Wave impedance inversion method, system, equipment and storage medium |
| CN114782744A (en) * | 2022-04-18 | 2022-07-22 | 清华大学 | Model processing method, device and equipment for reservoir prediction based on closed-loop network |
| CN116559949B (en) * | 2023-05-19 | 2024-07-12 | 北京宸宇金源科技有限公司 | Carbonate reservoir prediction method, system and equipment based on deep learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110231653A (en) * | 2019-04-24 | 2019-09-13 | 北京阳光杰科科技股份有限公司 | Construct the method and device of two-way constraint initial model |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10802171B2 (en) * | 2017-04-28 | 2020-10-13 | Pioneer Natural Resources Usa, Inc. | High resolution seismic data derived from pre-stack inversion and machine learning |
| CA3070479C (en) * | 2017-08-25 | 2023-01-17 | Exxonmobil Upstream Research Company | Automated seismic interpretation using fully convolutional neural networks |
| US10996372B2 (en) * | 2017-08-25 | 2021-05-04 | Exxonmobil Upstream Research Company | Geophysical inversion with convolutional neural networks |
| CN112213771B (en) * | 2019-07-10 | 2023-10-27 | 中国石油天然气股份有限公司 | Seismic wave impedance inversion method and device |
| CN111596366B (en) * | 2020-06-24 | 2021-07-30 | 厦门大学 | Wave impedance inversion method based on seismic signal optimization processing |
-
2021
- 2021-03-30 CN CN202110339626.5A patent/CN113109874B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110231653A (en) * | 2019-04-24 | 2019-09-13 | 北京阳光杰科科技股份有限公司 | Construct the method and device of two-way constraint initial model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113109874A (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113109874B (en) | Wave impedance inversion method using neural network and neural network system | |
| CN109709603B (en) | Seismic horizon identification and tracking method and system | |
| US11636240B2 (en) | Reservoir performance system | |
| Lorentzen et al. | History matching channelized reservoirs using the ensemble Kalman filter | |
| US10908308B1 (en) | System and method for building reservoir property models | |
| Kumar et al. | Ensemble-based assimilation of nonlinearly related dynamic data in reservoir models exhibiting non-Gaussian characteristics | |
| CN112576244B (en) | Porosity calculation method and device | |
| US11703608B2 (en) | Reservoir characterization using machine-learning techniques | |
| WO2021127186A1 (en) | Methods and systems for subsurface modeling employing ensemble machine learning prediction trained with data derived from at least one external model | |
| Li et al. | Simultaneous estimation of geologic and reservoir state variables within an ensemble-based multiple-point statistic framework | |
| Cornelio et al. | Residual learning to integrate neural network and physics-based models for improved production prediction in unconventional reservoirs | |
| WO2020256738A1 (en) | Field development planning based on deep reinforcement learning | |
| CN116971765A (en) | System and method for predicting and controlling hydrocarbon reservoir pressure | |
| CN114609667A (en) | Geophysical oil gas prediction method, equipment and storage medium | |
| Mohd Razak et al. | Rapid production forecasting with geologically-informed auto-regressive models: Application to Volve benchmark model | |
| CN113406695B (en) | Seismic inversion method and system based on interval velocity seismic geological model | |
| CN113642675B (en) | Underground rock stratum distribution imaging acquisition method, system, terminal and readable storage medium based on full waveform inversion and convolutional neural network | |
| Razak et al. | Physics-guided deep learning for improved production forecasting in unconventional reservoirs | |
| Razak et al. | Embedding physical flow functions into deep learning predictive models for improved production forecasting | |
| US20220206175A1 (en) | Training a machine learning system using hard and soft constraints | |
| Gervais et al. | Integration of saturation data in a history matching process based on adaptive local parameterization | |
| CN111949945B (en) | Method, device, equipment and system for determining total organic carbon content of source rock | |
| CN113608258A (en) | Self-consistent deep learning method for constructing high-resolution wave impedance inversion label | |
| Caers et al. | Statistical pattern recognition and geostatistical data integration | |
| US20240303398A1 (en) | Abnormal pressure detection using online bayesian linear regression |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |